{"source":"Quelle IA, le classement des modèles de langage","page":"https://quelleia.com/evolution/maths/","edition":"2026-W40","date":"2026-09-28","course":"maths","lecture":"Chaque marche est la sortie d'un modèle qui prend la tête de sa famille. Le score est celui d'aujourd'hui (édition du 28 septembre 2026), placé à la date de sortie : ce n'est pas un historique de mesures. Seuls les 81 modèles classés en maths comptent. La ligne des modèles ouverts commence en novembre 2025, à la sortie du premier modèle ouvert classé en maths. Chaque entrée est un modèle qui, à sa sortie, dépasse tous ceux de sa famille sortis avant lui. Le score est celui de l'édition : aucune mesure d'époque n'est rejouée.","depart":"2024-01-01","ecart_actuel":11,"modeles_classes":81,"fermes":[{"date_sortie":"2024-06-20","modele":"Claude 3.5 Sonnet","id":"claude-3-5-sonnet","editeur":"Anthropic","score":48.7,"fiche":"https://quelleia.com/modeles/claude-3-5-sonnet/"},{"date_sortie":"2024-12-17","modele":"o1","id":"o1","editeur":"OpenAI","score":69.3,"fiche":"https://quelleia.com/modeles/o1/"},{"date_sortie":"2025-01-31","modele":"o3-mini","id":"o3-mini","editeur":"OpenAI","score":71.4,"fiche":"https://quelleia.com/modeles/o3-mini/"},{"date_sortie":"2025-04-16","modele":"o4-mini","id":"o4-mini","editeur":"OpenAI","score":77.7,"fiche":"https://quelleia.com/modeles/o4-mini/"},{"date_sortie":"2025-08-07","modele":"GPT-5","id":"gpt-5","editeur":"OpenAI","score":82.6,"fiche":"https://quelleia.com/modeles/gpt-5/"},{"date_sortie":"2025-11-18","modele":"Gemini 3 Pro","id":"gemini-3-pro","editeur":"Google DeepMind","score":83.9,"fiche":"https://quelleia.com/modeles/gemini-3-pro/"},{"date_sortie":"2025-12-11","modele":"GPT-5.2","id":"gpt-5-2","editeur":"OpenAI","score":84.7,"fiche":"https://quelleia.com/modeles/gpt-5-2/"},{"date_sortie":"2026-02-05","modele":"Claude Opus 4.6","id":"claude-opus-4-6","editeur":"Anthropic","score":87.1,"fiche":"https://quelleia.com/modeles/claude-opus-4-6/"},{"date_sortie":"2026-03-05","modele":"GPT-5.4 Pro","id":"gpt-5-4-pro","editeur":"OpenAI","score":92.3,"fiche":"https://quelleia.com/modeles/gpt-5-4-pro/"},{"date_sortie":"2026-04-23","modele":"GPT-5.5 Pro","id":"gpt-5-5-pro","editeur":"OpenAI","score":94.9,"fiche":"https://quelleia.com/modeles/gpt-5-5-pro/"},{"date_sortie":"2026-06-09","modele":"Claude Fable 5","id":"claude-fable-5","editeur":"Anthropic","score":98.8,"fiche":"https://quelleia.com/modeles/claude-fable-5/"},{"date_sortie":"2026-09-01","modele":"Claude Fable 5.1","id":"claude-fable-5-1","editeur":"Anthropic","score":99.1,"fiche":"https://quelleia.com/modeles/claude-fable-5-1/"},{"date_sortie":"2026-09-03","modele":"GPT-6 Astra","id":"gpt-6-astra","editeur":"OpenAI","score":102,"fiche":"https://quelleia.com/modeles/gpt-6-astra/"}],"ouverts":[{"date_sortie":"2025-11-06","modele":"Kimi K2 Thinking","id":"kimi-k2-thinking","editeur":"Moonshot","score":76.1,"fiche":"https://quelleia.com/modeles/kimi-k2-thinking/"},{"date_sortie":"2025-12-01","modele":"DeepSeek-V3.2","id":"deepseek-v3-2","editeur":"DeepSeek","score":77.2,"fiche":"https://quelleia.com/modeles/deepseek-v3-2/"},{"date_sortie":"2026-01-27","modele":"Kimi K2.5","id":"kimi-k2-5","editeur":"Moonshot","score":79.5,"fiche":"https://quelleia.com/modeles/kimi-k2-5/"},{"date_sortie":"2026-04-07","modele":"GLM-5.1","id":"glm-5-1","editeur":"Z.ai (Zhipu AI)","score":81.6,"fiche":"https://quelleia.com/modeles/glm-5-1/"},{"date_sortie":"2026-04-20","modele":"Kimi K2.6","id":"kimi-k2-6","editeur":"Moonshot","score":83.7,"fiche":"https://quelleia.com/modeles/kimi-k2-6/"},{"date_sortie":"2026-06-16","modele":"GLM-5.2","id":"glm-5-2","editeur":"Z.ai (Zhipu AI)","score":85.5,"fiche":"https://quelleia.com/modeles/glm-5-2/"},{"date_sortie":"2026-07-16","modele":"Kimi K3","id":"kimi-k3","editeur":"Moonshot","score":91,"fiche":"https://quelleia.com/modeles/kimi-k3/"}],"licence":"Réutilisation libre en citant « Quelle IA, quelleia.com » et la date de l'édition."}